I have a flight collected with a DJI Mavic 3 pro in smart oblique mode. Over each line a forward tilted image is taken, followed by a backward tilted image, and then a ndar image. However, when importing the images in Pix4dmatic, the Yaw and pitch values are those of the aircraft, and not from the camera gimbal, so that the forward and backwadr tilte dimages are not properly oriented. I am having troubles wiht my photogrammetric solution, whihc cold be partly due to that. As anyone dealt with this or has knowledge on to best proceed?
Thanks for reporting Christophe,
This is unexpected, would you be able to share a dataset so we can have a look at what is going on?
1 Like
Dear Grégoire,
Here is what I get as solution. I feel it may be because the camera orientation are not set right from the ‘smart oblique’ mode capture of DJI (Mavi3 3E). The PPK geotagging did not show any errors…
What is the best way to share you the project? Via pix4d’s cloud?
Thanks
Christophe
Either a google drive link or a Pix4Dcloud project with images uploaded would be fine !
Thanks
Greg
Here goes the cloud link:
I have also tried with the mode ‘Oblique’, without much more success. The initial positions of the cameras appear to be fine, but after calibration I get a double layer plus a ‘ramp’ on one side. I have also tried to decrease the geolocation accuracy (2 cm for every image), without more success.
I will try the low texture options..
A quick recap:
Images were acquired with a Mavic3E in ‘smart oblique mode’ (the camera shifts the camera from looking forward at 20 degrees from Nadir, then 20 degrees backward, and then nadir.
The backward image is interpreted by pix4d as only a change in camera pitch, i.e. the camera yaw remains the same, but the pitch goes form positive to negative (backward), which is what the camera does. In DJI gimbal exif data, when the camera is facing backward, the pitch remains the same as forward (-70 relative to horizon), but the yaw is rotated 180 degrees… which is not the same as above.
1 Like
Thanks Christophe, we will have a look at your dataset. I will reach out in DM if needed.
Thanks for your patience.
Hello Grégoire,
I believe I found the problem… the two last flights of this multi-battery block were drone with digital zoom = 2… not sure how this happened. I picked it up will trying with agisoft metashape that gave me a warning of it. Agisoft is still able to process those images by changing the image size (or sensor size)… perhaps something to consider for pix4d?
Regards,
-Christophe
Thanks Christophe,
Thanks for sharing the data and glad you could solve the problem, we will look into how we can improve the situation!
Regards
Greg
Dear Christophe,
The issue with the zoom level is solved since PIX4Dmatic 1.78.0. In case you mixed images with different zoom levels, this is detected and multiple camera models with different internals are generated by the software. I have just successfully processed your project with default settings.
Thanks again for reporting the issue!
Davide
1 Like
Great! Tks
Envoyé à partir de Outlook pour Android
Dear Davide,
I re processed the data set. All cameras are calibrated now and the point cloud looks clean.
However in the report I get the warning that there is a 50% relative difference between initial and optimized camera parameters. Shouldn’t the initial focal length be changed for the zoomed camera?
Thanks,
(attachments)
20250317_Cipreses1rgb-quality_report_2025-07-11_09-51-45.pdf (1020 KB)
Following up on my previous thread: the point cloud densification fails, with the message:
[2025-07-11 17:00:18][07%CPU][23%RAM]Generate dense point cloud data Insufficient GPU resources to complete processing
[2025-07-11 17:00:18][07%CPU][23%RAM]Generate dense point cloud data Memory allocation error
The project has 4093 images (Mavic3E 20Mp RGB images). I have not had this problem before… is this related to the calibration?
-Christophe
Dear Christophe,
thanks a lot for reporting this, it is indeed a glitch that we need to address. The initial focal length is wrongly calculated for the camera corresponding to the images that have the digital zoom enabled. In any case, the algorithm figures out the correct focal length and the calibration is successful. The warning is not too much of a concern in this specific case.
We’ll work on this asap.